Adding `safetensors` variant of this model

Mix Trained marian model over 9 low resource languages of ALT dataset + 18K senetences from en-ja opus dataset.

Training in progress, step 20000

Training in progress, step 10000

initial commit